X
More Topics

AI, trust, and the art of conversation

Q&A: Splice Software CEO Tara Kelly on how artificial intelligence and data can improve how companies talk to their customers.
Written by Colin Barker, Contributor
splice-tara-kelly.jpg

Splice CEO Tara Kelly: "We want to create that connection between the person and the machine. And in order to do that, we can't just rely on the logical side of our brain. We have to engage the feeling side and this is where it gets really tricky."

Image: Splice Software

Splice Software uses big data and artificial intelligence to help companies to connect with their customers using voice, text, or email. Its technology creates automated messages by integrating data from multiple sources with scripted dialogues.

The company recently released a voice application for the Amazon Echo and is developing one for Google Home, with the aim of allowing companies to communicate with customers via these devices too.

Customers include insurance companies such as Majesco and Safety, as well as consumer customers such as La-Z-Boy and Grand Home Furnishings.

ZDNet spoke to Splice Software CEO Tara Kelly to find out more about the company and how AI is shaping communication between companies and their customers.

ZDNet: Tell me about the focus of Splice Software.

Kelly: It's all about personalisation to automation and specifically around all the areas where voice is involved. We believe it is all about the art and science of creating a beautiful connection for our customers, and we specialise in audio dialogues.

It's all around humanising that digital experience so that all of those platforms come together, thanks to innovation in things like AI and the IoT -- things like how we talk to our fridge and it's not science fiction. All of a sudden, it's here today.

Do you see a growing area as people become more conscious of what they should be able to do with technology, thanks to voice systems?

Yes, and I think that any form of adoption of a new technology is a very interesting curve. Automated calls with a digitised voice were actually available in the early '70s. And then we introduced that teddy bear that could talk in the early '80s [by the name of Teddy Ruxpin and he's back].

Back then people were making a go of it but it wasn't mainstream, or affecting how we lived our lives. Now we have moved on so much that we are looking at a situation where artificial intelligence is becoming a part of every decision we make.

It's about how that device can influence us. It's not all about access to databases and the intellectual side. It's about our ability to trust it. You know, if Siri sounded like garbage while she was giving you directions, you might doubt whether she was working.

But if she sounds really great, that's different, even though those things might not be related at all. It's very interesting when we get into the science of the human psyche and what is required for us to have trust.

So the crux of your technology is artificial intelligence?

I love the theory of artificial intelligence because I wonder how much of it is just insight and how much is true AI. That's because sometimes it may be something as simple as, say, knowing that in a certain region you might want to use a different dialect to create the connection.

As machines and humans try to interact, artificial intelligence is going to be heavily relied upon. And I think we have to be fair to the poor kid called artificial intelligence. I think it's in that very awkward teenage phase. It's not the bumbling toddler anymore but it still has a long way to go before it's perfect.

Is developing AI about making it more intuitive?

Yes, our goal is exactly that. We want to humanise that digital experience. We want to create that connection between the person and the machine. And in order to do that, we can't just rely on the logical side of our brain. We have to engage the feeling side and this is where it gets really tricky.

Humans are complex creatures. We're taking in multiple different cues to establish that sense of trust and security.

We are always looking for that emotional connection and tone. But that doesn't mean that everyone will need a long term relationship with that machine.

I'll give you an example.

If you walked into a shop to pick up a carton of milk and the lights were flickering and it looked dirty and you picked up the milk and walked to the clerk and he was rude -- then you are not likely to buy that milk. You can see that every transaction we make is based on a whole bunch of feedback before you trust and transact.

That is what we do as humans, but now you have these machines that are not feeding [in] as much of that rich visual experience that is needed to establish trust. You can see that things like audio become crucially important because it is one of the very few emotions that we can tap into when we use a machine.

This is why we are seeing an explosion in the use of video and why some of the greatest brands pump smells into their hotels or their stores because it creates that consistency and it builds that trust.

Then it's really all about getting to understand the user/consumer?

The beginning of it is really based on the quality of data and the dirty part of that is that most organisations don't have clean data. Part of the reason for that is that most organisations deploying have been through several acquisitions and have not even completely merged. So there is not a consistent base to work with.

You have this very important fact, the data. And you need to ensure that it is accurate and clean. The very first place that we have to start is to establish what level of data integrity we have. Nothing is worse that getting it wrong. Nothing is worse that only talking to only about two percent of your audience.

Then, at that point we are asking, what emotional state is that customer in? At what stage in that customer journey is this interaction likely to happen?

If we look at a Google Home or Amazon Alexa, where are they when we ask this particular question? Do we have enough information for this question to be accurate? As we look ahead, this is going to be very challenging and [this] is where AI is going to come in. Are you in the car or the kitchen? Where is that information going to be received? Questions like that.

The good news is that there's a lot going on in geomatics and in IP. So the first place we start is, how's your data?

If we look at voice input we have to start with a baseline: is it a neutral tone, a happy tone or an angry tone that reflects a disgruntled person? How can you tell? Interestingly, with disgruntled people the madder they get, the faster they speak.

This whole area is so complex. For example, how do you expect a voice input system to cope with regional accents?

You have to go to the community. You have to use things like open source. If organisations are going to lead in this area -- and there are only a couple who have been in speech for 15 or 30 years -- those organisations cannot keep this in silos. That's because all of their data points are against a certain targeted business/consumer that they have done business with and they now see as representative of our planet.

But right now we're terrible [with regional accents] because we architected it that way. We didn't collect enough data on a big enough group of people. One or two people just aren't a big enough dataset.

Do you have lots of times when customers are off-track and you try to steer them around?

Definitely. One of the gotchas to avoid is, if they have multiple languages in the home, don't assume that it's a language preference all the time. So if a mobile phone has been used with Spanish just that one time, don't start sending them everything in Spanish.

We see that often done wrong. It's like the system is built just to have a flag instead of offering that customer choices and preference.

Another big gotcha is, once you have collected some of that small data from your customer -- whether it's preference with the language, or preference in that they prefer one channel over the other -- you've got to make sure that they are using it.

So if they give you their consent to contact them somewhere, if they engage their Google Home device, you want to be talking to them there. They are not going to stay forever, they are not going to stay patient. Some companies slowly start an initiative, and then start to gather these things but have no real plans to deploy dialogs to their customers across those channels -- that can be a problem.

If you ask them for these details and then store this information, you have a responsibility to try and use it.

Another big faux pas is what I call the 'angry voice'. When a person is getting angry, that happy voice drives them crazy. You are trying to show them that you are this bright, successful, cheerful brand and this is not where this person is. If you want to engage this person through an automated agent you need to accommodate that. That's going to be a bit of a hurdle, and it will be for a while, to do that sort of mirror/match.

What is comes down to is: you need a lot of data, you need a lot of feedback, and you need some kind of artificial intelligence in those calls.

Read more about voice applications

Read more about customer relationship software

Editorial standards